Zhou et al. Cancer Cell Int (2021) 21:357 https://doi.org/10.1186/s12935-021-02008-5 Cancer Cell International

PRIMARY RESEARCH Open Access Prognostic signature composed of transcription factors accurately predicts the prognosis of gastric cancer patients Liqiang Zhou1, Zhiqing Chen2, You Wu1, Hao Lu1 and Lin Xin1*

Abstract Background: Transcription factors (TFs) are involved in important molecular biological processes of tumor cells and play an essential role in the occurrence and development of gastric cancer (GC). Methods: Combined The Cancer Genome Atlas Program and Genotype-Tissue Expression database to extract the expression of TFs in GC, analyzed the diferences, and weighted gene co-expression network analysis to extract TFs related to GC. The cohort including the training and validation cohort. Univariate Cox, least absolute contraction and selection operator (LASSO) regression, and multivariate Cox analysis was used for screening hub TFs to construct the prognostic signature in the training cohort. The Kaplan–Meier (K–M) and the receiver operating characteristic curve (ROC) was drawn to evaluate the predictive ability of the prognostic signature. A nomogram combining clinical infor- mation and prognostic signatures of TFs was constructed and its prediction accuracy was evaluated through various methods. The target genes of the hub TFs was predicted and enrichment analysis was performed to understand its molecular biological mechanism. Clinical samples and public data of GC was collected to verify its expression and prognosis. 5-Ethynyl-2′-deoxyuridine and Acridine Orange/Ethidium Bromide staining, fow cytometry and Western- Blot detection were used to analyze the efects of hub-TF ELK3 on the proliferation and apoptosis of gastric cancer in vitro. Results: A total of 511 misaligned TFs were obtained and 200 GC-related TFs were exposed from them. After sys- tematic analysis, a prognostic signature composed of 4 TFs (ZNF300, ELK3, SP6, MEF2B) were constructed. The KM and ROC curves demonstrated the good predictive ability in training, verifcation, and complete cohort. The areas under the ROC curve are respectively 0.737, 0.705, 0.700. The calibration chart verifed that the predictive ability of the nomogram constructed by combining the prognostic signature of TFs and clinical information was accurate, with a C-index of 0.714. Enriching the target genes of hub TFs showed that it plays an vital role in tumor progression, and its expression and prognostic verifcation were consistent with the previous analysis. Among them, ELK3 was proved in vitro, and downregulation of its expression inhibited the proliferation of gastric cancer cells, induced proliferation, and exerted anti-tumor efects. Conclusions: The 4-TFs prognostic signature accurately predicted the overall survival of GC, and ELK3 may be poten- tial therapeutic targets for GC

*Correspondence: [email protected]

1 Department of General Surgery, The Second Afliated Hospital of Nanchang University, 1 Minde Road, Donghu District, Nanchang 330006, Jiangxi, China Full list of author information is available at the end of the article

© The Author(s) 2021. This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat​ iveco​ ​ mmons.org/​ licen​ ses/​ by/4.​ 0/​ . The Creative Commons Public Domain Dedication waiver (http://creat​ iveco​ mmons.​ org/​ publi​ cdoma​ in/​ ​ zero/1.0/​ ) applies to the data made available in this article, unless otherwise stated in a credit line to the data. Zhou et al. Cancer Cell Int (2021) 21:357 Page 2 of 17

Keywords: Gastric cancer, Transcription factors, Prognostic signature, Nomogram, ELK3

Introduction prognosis of these hub TFs. We also jointly established Gastric cancer is a common malignant tumor of the a nomogram with the prognostic signature and clinical digestive tract. Its incidence ranks ffth among all malig- information and used a variety of methods to verify its nant tumors, and its mortality ranks fourth among can- predictive performance, which is helpful for clinicians to cer-related deaths [1]. Due to factors such as economic make decisions. More importantly, we identifed a novel level and lifestyle, China has become an area with a high biomarker for gastric cancer and found it regulated cell incidence of gastric cancer. In 2020, the new cases and cycle and proliferation in vitro. deaths of gastric cancer in China accounted for 62.3% and 51.4% of the global total [1]. Te clinical manifesta- Materials and methods tions of gastric cancer have no signifcant specifcity, Data processing similar to the manifestations of non-malignant gastroin- Te names of transcription factor gene were obtained testinal diseases, with insidious onset, rapid progress, dif- from the Human Transcription Database website (http://​ fculty in early diagnosis, and mostly in the middle and bioin​fo.​life.​hust.​edu.​cn/​Human​TFDB#​!/) [7]. Gastric late stages of diagnosis, making most cases lose the best cancer RNA-seq data and clinical information were opportunity for surgery. Te clinical manifestations of obtained from the TCGA website. We standardized the advanced gastric cancer are mostly merged by invasion downloaded FPKM data of gastric cancer and converted of adjacent tissues or organs, lymph node metastasis in it into TPM data, and combined the TPM data of nor- the abdominal cavity and organ metastasis. At present, mal gastric mucosa downloaded on the GTEx website by the main treatment for patients with gastric cancer in removing the batch efect. After extracting the transcrip- this stage is chemotherapy. However, this treatment has tion factor expression data, using the “limma” R pack- large side efects, low quality of life of patients, and short age [8], False Discovery Rate (FDR) < 0.05 and |log2Fold survival period [2]. Terefore, seeking a stable and efec- Change (FC)| > 1.0 as screening conditions to identify dif- tive diagnostic index for gastric cancer and solving the ferentially expressed transcription factors. Te “ggplot” R problems encountered in clinical treatment is the focus package was used to draw volcano maps and heat maps of gastric cancer research. to visualize diferentially expressed transcription factors. Transcription factors are a group of protein mol- ecules that can specifcally bind to a specifc sequence Weighted gene co‑expression network analysis (WGCNA) upstream of the 5′end of a gene to ensure that the tar- To search the TFs that are highly correlated with gas- get gene is expressed at a specifc time and space with a tric cancer, the DETFs obtained were analyzed using specifc strength. Teir function is to regulate, turn on the “WGCNA” R package [9]. We frstly used the Pear- and turn of genes to guarantee that the correct number son correlation coefcient of gene pairs to establish an of genes expressed in the correct cell at the correct time unsupervised co-expression relationship based on the throughout the entire life of the cell and organism [3]. adjacency matrix of connection strength. Ten we used Among genetic factors, TFs play a vital role in the most topological overlap matrix analysis to cluster the adja- important cellular processes, such as cell development, cency matrix of the gene expression data of gastric can- response to internal and external environmental changes, cer patients. Finally, the dynamic tree-cutting algorithm cell cycle control, and carcinogenesis [4]. TFs are the was applied to the tree diagram for module identifcation, drivers of tumor initiation and disease progression, and the minimum size of the module gene number was set their remarkable diversity and efectiveness making their to 30, and the cutting height was 0.90. Using the expres- attractive prognostic and therapeutic targets for cancer sion data of each co-expression module in all samples [5, 6]. to execute module feature genes (MEs) as the frst main In this study, we combined Te Cancer Genome Atlas component. Te candidate TFs in the modules with the (TCGA) and Genotype-Tissue Expression (GTEx) data- highest and lowest correlation with GC were extracted bases on the gene expression and corresponding clini- for further analysis. cal information of gastric cancer. Several TFs related to the overall prognosis were identifed, and a prognostic Construction and verifcation of prognostic signatures signature of TFs was developed to predict the overall Univariate Cox regression analysis screened out TFs survival of gastric cancer patients. Clinical samples and related to the overall prognosis from DETFs. Te LASSO public databases was used to verify the expression and regression was further analyzed and the collinearity was Zhou et al. Cancer Cell Int (2021) 21:357 Page 3 of 17

removed to obtain TFs that were signifcantly related Transcription target gene prediction and enrichment to the prognosis [10]. Ten, the samples in TCGA were analysis randomly divided into the training and verifcation Using the Gene Transcription Regulation Database cohort. In the training cohort, multivariate Cox regres- (GTRD) database (http://​gtrd.​biouml.​org), we analyzed sion was used to construct a prognostic signature in TFs that within 2000 kb upstream and downstream of the with signifcant prognosis. Te hub TFs with non-zero transcription start site, SiteCount ≥ 10 is the target gene coefcients were selected to calculate the risk score. bound by hub TFs [13]. Te “org.Hs.eg.db” R package was Te prognostic risk score of each patient applies to the used to perform GO and KEGG function enrichment following formula: risk score = expression level of TF analysis, among which items that meeting p value < 0.05 1 × Cof1 + expression level of TF 2 × Cof 2 + … + Expres- and q value < 0.05 are signifcant, to explore the potential sion level of TF x × Cof x, where Cof represents the value function of hub TF in gastric cancer. of each TF Regression coefcients. Te median risk score was used as the cut-of value to divide STAD patients in Hub TFs expression and prognosis characteristics the training cohort into high-risk and low-risk groups. and verifcation Using the same formula and the same cut-of value in We frst analyzed the relationship between hub TFs and the verifcation queue. Drawing the K–M survival curve, various clinicopathological characteristics, and explored and using the log-rank test to evaluate the diference in their expression characteristics. Ten, 10 pairs of gas- survival between the high and low-risk groups. Te sen- tric cancer clinical surgical specimens were collected. sitivity and specifcity of the prognostic signature was All procedures were approved by the patient’s informed calculated by the 5-year ROC curve [11]. Univariate and consent and the ethics committee of the Second Afli- multivariate Cox regression analyses were performed to ated Hospital of Nanchang University. After the sample confrm whether the prognostic model of TFs is an inde- was homogenized, the total RNA was extracted with Tri- pendent prognostic factor compared with the clinical zol (Termo Fisher, USA), and the RNA obtained was prognosis. Also, we merged the training and validation reverse transcribed using the reverse transcription kit cohorts and used the same methods for analysis. RR047A (Takara, Japan). ACTB was used as the internal reference gene, and the mRNA expression of hub TFs was Pathway analysis analyzed by rt-PCR using the RR820 kit (Takara, Japan) To explore the molecular mechanism diferences between on the 7900-HT system (Termo Fisher, USA). Te prim- high and low-risk groups. We performed principal com- ers were all synthesized by Shanghai Shenggong, see the ponent analysis to understand the diference between attached table for details. In addition, the HPA database high and low-risk groups. Using Gene Set Enrichment was used to analyze the protein expression of hub TFs Analysis (GSEA, version 4.0) based on the molecu- [14]. Finally, the prognosis of hub TFs in GSE51105 was lar signature database (Molecular Signatures Database, verifed on Kaplan–Meier Plotter. MSigDB) to provide gene enrichment analysis for the high-risk group and low-risk group (|NSE| > 1, FDR < 0.05 Cell line selection and transfection is considered as statistical learning meaning. Using this Based on the above analysis results, we speculated method to fnd the diferences in tumor pathways and that ELK3 is a potential new biomarker for gastric can- mechanisms between high and low-risk groups. cer. In order to verify the function of ELK3, download the GSE146361 microarray data from Gene Expres- sion Omnibus to analyze the expression of ELK3 in Nomogram construction and verifcation gastric cancer cell lines, and obtain a cell line with high Using the “rms” R package to build a prognostic nomo- expression of ELK3. Furthermore, according to lipo3000 gram for STAD patients to predict the probability of (Termo Fisher Scientifc, USA) instructions, RNA inter- survival in 1–5 years. Age, Gender, Radiation therapy, ference technology was used to inhibit the expression of Pharmaceutical therapy, Pathological stage, pathologi- ELK3 in cells. Te siRNA used were all synthesized by cal T stage, pathological N stage, pathological M stage, Sangon Biotech (Shanghai, China). Te sequence is as and risk score are independent parameters that form the follows, SiScr: Sense 5′-UUC​UCC​GAA​CGU​GUC​ACG​ nomogram. Using C index and calibration curve to cal- UTT-3′, Antisense 5′-ACG​UGA​CAC​GUU​CGG​AGA​ culate the discrimination and calibration of nomogram ATT-3′; siELK3-1: Sense 5′-CCU​GCG​AUA​CUA​UUA​ prediction and true survival rate [12]. By quantifying the UGA​CAATT-3′, Antisense 5′-UUG​UCA​UAA​UAG​UAU​ net income under diferent threshold probabilities in the CGC​AGGTT-3′; siELK3-2: Sense 5′-UGG​AUC​AGA​ nomogram, a decision curve analysis was carried out to AACAUG​ AGC​ AUUTT-3​ ′, Antisense 5′-AAUGCU​ CAU​ ​ determine the clinical validity of the nomogram. Zhou et al. Cancer Cell Int (2021) 21:357 Page 4 of 17

GUU​UCU​GAU​CCATT-3′; siELK3-3: Sense 5′-AUC​ temperature. Te band was detected using Super ECL AGG​UUU​GUG​ACC​AAU​AAATT-3′, Antisense 5′-UUU​ Plus (US EVERBRIGHT, China). Te protein expres- AUU​GGU​CAC​AAA​CCU​GAUTT-3′. Tree days after sion results are expressed relative to the GAPDH band transfection, the expression changes of ELK3 were ana- density. lyzed by rt-PCR. Results Cell proliferation assay Identifcation of diferentially expressed and gastric Te proliferation of gastric cancer cells was evaluated cancer‑related transcription factors by 5-Ethynyl-2′-deoxyuridine (EDU) Cell proliferation Te analysis process of this study was shown in Fig. 1. detection. Staining according to the instructions of the Tere are a total of 375 gastric cancer samples and 32 EDU commercial kit (US EVERBRIGHT, Suzhou, China), normal samples in the gastric adenocarcinoma (STAD) and using a fuorescence microscope (Olympus, Japan) cohort of the TCGA database, and there are 359 normal to perform EDU measurement on the treated cells. Per- samples in the GTEx database. Te clinical information forming PI single staining on the cells according to the of TCGA-STAD was shown in Additional fle 1: Table S1. cell cycle kit (US EVERBRIGHT, Suzhou, China), instruc- We extracted the expression data of 1,634 transcription tions, using Becton Dickinson FACS calibur instrument factors and identifed 284 up-regulated and 227 down- to analyze the cell cycle distribution, and analyze the regulated transcription factors based on the screening efect of ELK3 on cell proliferation. conditions (Additional fle 1: Table S2). Using volcano plot (Fig. 2A) and heat map (Fig. 2B) to visually display. Cell apoptosis detection Using WGCNA to further analyze these 511 transcrip- In order to analyze the efect of inhibiting ELK3 on the tion factors. We frst determined whether there are out- apoptosis of gastric cancer cells, frst stained with Acri- liers in each sample, and then performed hierarchical dine Orange/Ethidium Bromide(AO/EB) Kit (Sangon clustering. In the WGCNA analysis, we chose the soft Biotech, Shanghai, China), and analyzed the number of threshold capability to determine the relative balance of apoptosis of gastric cancer cells after downregulating scale independence and mean connectivity. As shown ELK3. Further, using Annexin V-APC Apoptosis Detec- in Fig. 3A, power = 15 can be used as the power value of tion Kit (US EVERBRIGHT, Suzhou, China) to detect the soft threshold. Ten, based on the input TFs, through cells in the early and late stages of apoptosis. Te cells average linkage hierarchical clustering, a total of 10 mod- were processed according to the instructions, collected ules were generated (Fig. 3B). After calculating the cor- and analyzed in a Becton Dickinson FACS calibur instru- relation MS of the shape of each module (Fig. 3C), the ment. Te cells that were positive for Annexin V-APC MEbrown module containing 49 TFs was considered to and PI were counted. be the most relevant to gastric cancer, and the MEtur- quoise containing 151 TFs was considered the least rel- Western Blot analysis evant to gastric cancer. Te specifc gene names are in Te cells were lysed in RIPA (Solarbio, China) containing Additional fle 1: Table S3. protease inhibitors (Boster, China) for 20 min on ice. Te bicinchoninic acid protein content kit (Solarbio, China) Construction and verifcation of TFs prognostic signature to determine protein concentration. 40 μg total protein Univariate Cox regression analysis identifed 8 TFs per well was separated on 10% polyacrylamide gels and related to the overall prognosis, among which dangerous transferred to polyvinylidene fuoride (PVDF) mem- TFs were shown in red and protective TFs were shown in brane (Merck, Germany). Te membrane was blocked green (Table 1). Te LASSO regression analysis was per- with 5% BSA for 1 h at room temperature. Te PVDF formed on these TFs to further determine the prognos- membrane was combined with GAPDH (Proteintech, tic signifcantly related TFs, which are ZNF300, ELK3, USA, Cat No. 60004-1-Ig), PCNA (ABclonal, China, Cat SP6, ZNF564, MEF2B, FOXS1 (Fig. 4). Subsequently, we No. A0264), P21 (ABclonal, China, Cat No. A1483), P16 divided the TCGA-STAD queue into a training cohort (ABclonal, China, Cat No. A0262), B-cell lymphoma/leu- and a verifcation cohort. Based on these 4 TFs, multi- kemia-2 (Proteintech, USA, Cat No. 12789-1-AP), BCL2 variate Cox regression analysis was used in the training Associated X (Proteintech, USA, Cat No. 50599-2-Ig), cohort to further construct the prognostic signature, and Caspase-3 (Proteintech, USA, Cat No. 66470-2-Ig)was fnally, 4 TFs were obtained. Te relative regression coef- incubated overnight. After washing with Tris-bufered fcients were shown in Table 1. saline Tween, the membrane was probed with horse- By calculating the risk score of each patient, using radish peroxidase-conjugated goat anti-rabbit IgG or the median as the threshold, they were divided into goat anti-mouse IgG (Boster, China) for 1 h at room high-risk and low-risk groups. Kaplan–Meier (KM) Zhou et al. Cancer Cell Int (2021) 21:357 Page 5 of 17

Fig. 1 Flow chart of this research

Fig. 2 The identifycation of 494 diferentially expressed TFs combined with TCGA and GTEx databases. A Volcano plot. B Heat map

survival analysis showed (Fig. 5A) that the high-risk that the prognostic signature has moderate predic- group had a lower survival rate (P = 1.772e−05). tive sensitivity and specifcity (Fig. 5B). We performed Besides, the 5-year receiver operating characteris- univariate and multivariate Cox regression analysis tic curve (ROC) was drawn and the area under the to assess the prognostic value of risk scores. Univari- curve (AUC) was calculated to be 0.737, indicated ate Cox regression showed (Fig. 5C) Pathologic stage Zhou et al. Cancer Cell Int (2021) 21:357 Page 6 of 17

Fig. 3 WGCNA identifed 200 TFs closely related to gastric cancer. A Analyze the best Power value. B Dendrogram of all clusters expressing TFs based on the diference metric (1-TOM). C The heat map shows the correlation between each module and gastric cancer

Similarly, we verifed the prognostic signature in the [HR = 1.763, 95% CI (1.274–2.440), P < 0.001], T stage verifcation cohort, and the K–M curve survival analy- [HR = 1.500, 95% CI (1.083–2.079), P = 0.015], M stage sis showed (Fig. 5E) that the prognosis of the high-risk [HR = 2.4422, 95% CI (1.097–5.349), P = 0.029], N stage group was worse (P 2.651e 02). Te 5-year AUC was [HR = 1.352, 95% CI (1.075–1.701), P = 0.010] and risk = − 0.705, showing good specifcity and sensitivity (Fig. 5F). score [HR = 1.931, 95% CI (1.373–2.714), P < 0.001]. Multivariate Cox regression analysis showed (Fig. 5D) Univariate and multivariate Cox regression analysis showed (Fig. 5G, H), Age [HR 1.074, 95% CI (1.045– Gender [HR = 2.043, 95% CI (1.088–3.833), P = 0.026], = 1.103), P < 0.001], Gender [HR 2.135, 95% CI (1.235– Radiation therapy [HR = 0.314, 95% CI (0.314–0.836), = 3.688), P 0.007], Tumor grade [HR 1.870, 95% CI P = 0.020], And risk score [HR = 2.237, 95% CI (1.505– = = 3.325), P < 0.001] were independent prognostic factors. (1.118–3.128), P = 0.017], M stage [HR = 2.997, 95% CI Zhou et al. Cancer Cell Int (2021) 21:357 Page 7 of 17

Table 1 Univariate and multivariate Cox regression screening prognostic-related transcription factors Gene ID Univariate Cox regression Multivariate Cox regression Coef. HR 95% CI P value HR 95% CI P value

ZNF300 1.248 1.248–1.523 0.030 1.381 1.025–1.860 0.034 0.323 ELK3 1.501 1.177–1.913 0.001 1.371 0.937–2.006 0.104 0.315 SP6 0.802 0.701–0.919 0.001 0.791 0.641–0.977 0.029 0.235 − MEF2B 0.564 0.340–0.935 0.026 0.503 0.245–1.033 0.061 0.686 − FOXS1 1.232 1.057–1.437 0.008 – – – – E2F2 0.796 0.668–0.948 0.011 – – – – KLF9 1.230 1.061–1.428 0.006 – – – – ZNF564 0.403 0.230–0.707 0.002 – – – –

HR hazard ratio, CI confdence interval, Coef. coefcients

Fig. 4 Lasso regression removes collinearity to identify 6 candidate TFs

(1.029–8.731), P = 0.044], N stage [HR = 1.680, 95% CI P < 0.001], Gender [HR = 1.596, 95% CI (1.092–2.334)), (1.211–2.330), P = 0.002], risk value [HR = 1.890, 95% P = 0.016], Radiation therapy [HR = 0.389, 95% CI CI (1.251–2.856), P = 0.003] are independent prognostic (0.213–0.710), P = 0.002], N stage [HR = 1.288, 95% CI factors. (1.023–1.622), P = 0.031] these factors, the risk score Also, we analyzed the entire TCGA-STAD cohort. Te [HR = 1.831, 95% CI (1.408–2.381), P < 0.001] has better scatter chart showed the distribution of risk scores and predictive ability (Fig. 6D, E). the correlation between risk scores and survival data. Patients in the high-risk group had higher mortality and Pathway analysis lower survival time (Fig. 6A). Te K-M curve survival First, the principal component analysis (PCA) showed analysis showed (Fig. 6B) that the low-risk group had a that there were signifcant diferences between the high higher survival rate (P = 4.520e−06). Te 5-year AUC and low-risk groups (Fig. 7A). Pathway analysis using value is 0.700, which is not signifcantly diferent from GSEA (Fig. 7B) showed that gastric cancer samples in the the training set and the validation set (Fig. 6C). Univari- high-risk group were mainly enriched in Angiogenesis, ate and multivariate Cox regression analysis showed that Epithelial Mesenchymal Transition, Hedgehog signaling, compared with Age [HR = 1.037, 95% CI (1.018–1.057), Hypoxia, IL2/STAT5 signaling, Infammatory Response, Zhou et al. Cancer Cell Int (2021) 21:357 Page 8 of 17

Fig. 5 Construction and verifcation of prognostic signatures in training and verifcation cohort. A, E K–M curve analysis of survival diferences between high and low-risk groups. B, F 5-year ROC curve analysis of the sensitivity and specifcity of the prognostic signature. C, G Univariate Cox regression analysis of the relationship between prognosis model, clinical case characteristics, and prognosis. D, H Multivariate Cox regression analysis whether the prognostic model and clinical case characteristics are independent prognostic factors

KRAS signaling up, NOTCH signaling, TGF-BETA sign- stage, RiskScore. Te total points of each patient pro- aling, NFKB/TNFA signaling. Tese pathways play an vided the estimated 1–5 year survival times (Fig. 8A). important role in the occurrence and development of Te C-index of this nomogram is 0.714. As shown by tumors, suggesting that patients with high-risk gastric the calibration chart, the actual 5-year survival rate cancer have a higher degree of tumor malignancy. matches well with the 5-year survival rate predicted by the calibration chart (Fig. 8B). Decision curve dis- play (Fig. 8C), if the threshold probability of a patient Nomogram construction and verifcation and a doctor is > 14 and < 67%, respectively, using this Te nomogram is an efective tool that integrate multi- nomogram to predict gastric cancer patients prognosis ple risk factors for clinical applications. We established more beneft than the scheme. Within this range, the a nomogram of the overall prognosis for 1–5 years in net beneft was comparable with several overlaps, based the TCGA-STAD cohort. Te model integrates Age, on the nomogram. Gender, Radiation therapy, Pharmaceutical therapy, Tumor grade, Pathologic stage, T stage, M stage, N Zhou et al. Cancer Cell Int (2021) 21:357 Page 9 of 17

Fig. 6 Overall analysis of prognostic signatures in the TCGA-STAD cohort. A Heat maps and scatters plots show that the high-risk group has higher mortality and shorter survival time for gastric cancer patients. B The K–M curve shows that the high-risk group has a worse prognosis. C The 5-year ROC curve shows that the prognostic model has good predictive performance. D, E Univariate and multivariate Cox regression analysis shows that the prognostic model is an independent prognostic factor Zhou et al. Cancer Cell Int (2021) 21:357 Page 10 of 17

Fig. 7 Analyzes the diference between the high and low-risk groups. A Principal component analysis. B GSEA shows that the high-risk group gastric cancer samples are enriched in pathways closely related to the tumor

Hub TFs target gene prediction and enrichment analysis pairs of clinical samples. Te primer sequences are in Te GTRD is used to predict the target genes of 4 hub Table 2. Te results suggested (Fig. 10B) that the expres- TFs. Among them, there are 623 eligible target genes for sion of ELK3 and SP6 is increased in gastric cancer, and ELK3, 449 for SP6, 1569 for MEF2B, and 89 for ZNF300 the expression of ZNF300 and MEF2B is down-regulated (Additional fle 1: Table S4). Perform gene ontology in gastric cancer. In addition, using Te Human Protein (GO) and “Kyoto Encyclopedia of Genes and Genomes” Atlas (HPA) to analyze the protein expression of hub TFs, (KEGG) analysis on these target genes. GO showed ELK3 immunohistochemical staining intensity in normal (Fig. 9A) that biological processes were enriched in regu- tissues is lower than gastric cancer tissues, while ZNF300 lation of GTPase activity, regulation of cell morphogene- and MEF2B are higher than gastric cancer tissues sis, Ras protein signal transduction, etc., cell components (Fig. 10C). Kaplan–Meier Plotter showed that the expres- were enriched in neuron to neuron synapse, focal adhe- sions of ELK3 (P = 0.014) and ZNF300 (P = 0.110) in the sion, cell-substrate junction, etc., and molecular func- GSE51105 data set were associated with poor progno- tions were enriched in guanyl-nucleotide exchange factor sis, and low expression of SP6 (P = 0.110) and MEF2B activity, small GTPase binding, Ras GTPase binding. For (P = 0.100) suggested a better prognosis (Fig. 10D). KEGG (Fig. 9B), target genes were mainly enriched in important signals involved in tumorigenesis and develop- Inhibition of ELK3 can inhibit the proliferation of gastric ment, such as MAPK signaling pathway, Wnt signaling cancer cells and induce apoptosis pathway, Autophagy, and Rap1 signaling pathway. Based on the above analysis results, we found that ELK3 is not only highly expressed in gastric cancer, but also Hub TFs expression and prognosis characteristics related to poor prognosis. Terefore, further analysis of and verifcation the role played by ELK3 in gastric cancer cells. First, we We analyzed the relationship between hub TFs and clin- analyzed the expression of ELK3 in 27 cell lines from icopathological characteristics, and the results showed the GSE146361 microarray, and the results indicated that the expression of SP6 was related to the grade and that the expression of Hs746t was the highest (Fig. 11A). age of gastric cancer, and the expression of ELK3 was Furthermore, using Hs746t as an in vitro verifcation related to the grade, and its expression increased with experimental cell line, three siRNAs were used to inhibit the depth of tumor invasion(Fig. 10A). Ten, we used the expression of its ELK3. Western-blot showed that rt-PCR to verify the mRNA expression of hub TFs in 10 all three siRNAs had good efects (Fig.11B). We chose Zhou et al. Cancer Cell Int (2021) 21:357 Page 11 of 17

Fig. 8 Construction and verifcation of nomogram. A Combining clinicopathological characteristics and prognostic signatures to construct a nomogram to predict the 1–5 year survival rate of gastric cancer patients. B The 5-year calibration chart verifes the predictive ability of the nomogram. C 5-year decision curve analysis of clinical beneft rate Zhou et al. Cancer Cell Int (2021) 21:357 Page 12 of 17

Fig. 9 Enrichment analysis of hub TFs target genes. A Gene ontology, B Kyoto Encyclopedia of Genes and Genomes

the second siRNA for further experiments, and rt-PCR activating the VEGFA promoter [16]. Insulin gene showed that it can inhibit the RNA expression of ELK3 in enhancer protein 1 (ISL1) promotes glycolysis and tum- cells (Fig. 11C). origenesis in GC through transcriptional regulation of In order to verify the efect of ELK3 on the proliferation GLUT4 [17]. β-catenin can regulate the expression of of gastric cancer, we performed EDU staining. Te results PD-L1 to induce immune escape in gastric cancer [18]. showed that the proliferation of Hs746t cells decreased Tese evidence indicated that transcription factors play after ELK3 was inhibited (Fig. 12A). Cell cycle analysis an important role in gastric cancer. In-depth explora- indicated that after inhibiting ELK3, the proportion of tion of the potential molecular functions of transcription cells in G1 phase was increased, while that in S phase was factors and using them as therapeutic targets has great decreased, and the cell proliferation ability was weak- prospects. ened (Fig. 12B). Western-blot showed that the expres- We jointly analyzed the data of TCGA and GTEx with sion of cell proliferation-related proteins PCNA, P21, P16 the diferences in the expression of transcription factors decreased with the down-regulation of ELK3 (Fig. 12C). in gastric cancer as a whole, and identifed transcrip- Finally, we analyzed the efect of inhibiting ELK3 on cell tion factors that are closely related to the prognosis. On apoptosis. After AO/EB staining, the expression of ELK3 this basis, we constructed a prognostic risk proportional was decreased and the number of apoptosis of Hs746t model and verifed its good predictive performance. was increased (Fig. 13A). Flow cytometry detection Trough systematic analysis, we found that the high-risk showed that the rate of apoptosis was negatively corre- group has a worse prognosis. GSEA further explained lated with the expression of ELK3 (Fig. 13B). Western- that the high-risk group was mainly enriched in Angio- blot showed that after ELK3 was inhibited, the expression genesis, Epithelial Mesenchymal Transition (EMT), of anti-apoptotic protein Bcl-2 decreased, and the Hedgehog signaling, Hypoxia, IL2/STAT5 signaling, expression of pro-apoptotic proteins Bax and Caspase-3 Infammatory response, KRAS signaling up, Notch sign- increased (Fig. 13C). aling, TGF-beta signaling, NF-κB/TNFA signaling. Tese signals play an important role in tumors and participate Discussion in the occurrence and development of tumors. EMT is Since the discovery of transcription factors in 1961, there one of the key mechanisms of cell morphological plas- has been increasing evidence that transcription factors ticity changes in embryonic development and tumor are key drivers of many diseases, including cancer [15]. metastasis. It is essential in tumor invasion and metas- In gastric cancer, there have been multiple reports show- tasis progression. Te process was mainly manifested by ing that transcription factors play an important role. For tumor epithelial cells losing epithelial cell polarity under example, cyclic AMP response element binding protein specifc conditions. Te contacts between the surround- 3-like 4 (CREB3L4) promotes the progression of gastric ing cells and the matrix is reduced, the adhesion between tumors and endothelial angiogenesis by transcriptionally the cells is reduced, the interstitial characteristics are Zhou et al. Cancer Cell Int (2021) 21:357 Page 13 of 17

Fig. 10 Expression and prognostic verifcation of hub TFs. A rt-PCR analyzes the expression of hub TFs in surgical samples. B HPA database analyzes the protein expression level of hub TFs. C Analyze the prognosis of hub TFs in GSE51105

obtained, and the cell phenotype is changed. After this reach distal tissues or organs to form new tumor metas- process, the tumor cells break through the basement tases [19]. Among them, the mechanisms that trigger membrane, causing the adhesion between the cells or the EMT in tumors include: transforming growth factor-β matrix to decrease or disappear, migration and invasive- (TGF-β), Wnt, Notch, and Hedgehog signals. Tis indi- ness increase, and enter the lymph and blood vessels to cated that the high-risk group of gastric cancer patients Zhou et al. Cancer Cell Int (2021) 21:357 Page 14 of 17

Table 2 Hub transcription factors primer sequence

Gene ID Forward primer sequence (5′–3′) Reverse primer sequence (5′–3′)

ZNF300 GAG​TAA​CCT​TCA​CAA​CTC​CCAG​ ATG​CCT​CAG​TCA​CTG​TTT​TGC​ ELK3 GAG​AGT​GCA ATC​ACG​CTGTG​ GTT​CGA​GGT​CCA​GCA​GAT​CAA​ SP6 CAG​CCT​CTC​CAA​ACT​TAC​CAG​ AGG​TCC​TCG​CAG​GTT​ACC​C MEF2B ATG​GAC​CGT​GTG​CTG​CTG​AAGT​ TCCGAA​ ACT​ TCT​ CTC​ CTG​ GCTC​ ​ ACTB CAC​CAT​TGG​CAA​TGA​GCG​GTTC​ AGG​TCT​TTG​CGG​ATG​TCC​ACGT​

Fig. 11 Cell line selection and verifcation of ELK3 inhibition efciency. A The expression of ELK3 in each gastric cancer cell line in the GSE146361 microarray. B The expression of ELK3 in each gastric cancer cell line in the GSE146361 microarray. C rt-PCR analysis inhibits the transfection efciency of ELK3

progresses more rapidly and has a higher degree of in the development of muscles, heart, bones, blood ves- malignancy [20]. sels, and the immune system. However, studies have We also identifed 4 hub TFs, MEF2B, SP6, ZNF300, found that MEF2B can activate the β-catenin pathway to ELK3, and predicted their target genes. Te enrichment induce lung cancer cell invasion [21]. Tis may be due to analysis of these target genes showed that the target genes diferences in gene expression and functions in diferent were mainly enriched in important signals involved in microenvironments. SP6 belongs to the transcription fac- tumorigenesis and development, such as MAPK sign- tor family, which contains three classic zinc fnger DNA aling pathway, Wnt signaling pathway, Autophagy, and binding domains, which are composed of two cysteines Rap1 signaling pathway. MEF2B is a member of the MEF2 and two histidines (C­ 2H2 motif) tetrahedral coordinated family of proteins and is a transcription factor involved zinc atoms, these transcription factors bind to GC-rich Zhou et al. Cancer Cell Int (2021) 21:357 Page 15 of 17

Fig. 12 The cell proliferation ability is weakened after ELK3 is inhibited. A Inhibition of ELK3 expression cell proliferation decreased. B After inhibiting ELK3, gastric cancer cells were blocked in G1 phase. C Cell proliferation-related protein after ELK3 expression down-regulation

sequences and related GT and CACCC boxes [22]. At protein 1, which can bind to a specifc DNA sequence present, there is no experimental study on the mecha- rich in purine GGA core sequences and regulate the nism of SP6 in afecting tumor progression. In this study, expression of a variety of genes including proto-onco- MEF2B and SP6 were considered protective genes, while genes [27]. Under basic conditions, ELK3 is a transcrip- ELK3 and ZNF300 were considered oncogenes. ZNF300 tional repressor, but it can be activated by RAS/ERK is a novel KRAB/C2H2 gene encoding 68kD ZFP, and its signals and mitogen-activated protein kinase (Mitogen- KRAB domain exhibits transcriptional repressive activ- Activated Protein Kinase, MAPK) pathways to turn it ity [23]. What is interesting is that Endogenous ZNF300 into a transcription activator [28, 29]. In recent years, binds directly to the IL2RB gene promoter and potentially ELK3 has been proven to play an important role in the activates its expression [24]. Reports in tumors indicated occurrence and development of breast cancer, liver can- that ZNF300 can promote the progression of cancer cells cer, lung cancer, and other malignant tumors [30–33]. In by activating NF-κB and MAPK pathways to induce tumor prostate cancer studies, it has been shown that inhibition cell proliferation, invasion, and drug resistance [25, 26]. of ELK3 can promote cycle arrest and apoptosis of tumor It is worth noting that although ZNF300 is associated cells [34]. In the reports of breast cancer and colorectal with poor prognosis, it is low expressed in tumors. Te cancer, ELK3 is closely related to chemotherapy resist- expression of ELK3 in gastric cancer is positively cor- ance, and down-regulating its expression can promote related with poor prognosis. ELK3 (also known as Net, chemotherapy sensitivity [35, 36].In addition, ELK3 is SAP-2, or ERP) is a member of the ETS transcription also involved in TGF-β signaling to promote tumor cells factor family and is located on chromosome 12q23.1. to undergo epithelial-mesenchymal transition [37, 38]. Te ELK3 protein often forms a ternary complex tran- In gastric cancer, ELK3 has no experimental studies to scription factor together with serum response accessory confrm its function. Our in vitro studies have shown Zhou et al. Cancer Cell Int (2021) 21:357 Page 16 of 17

Fig. 13 Inhibit ELK3 expression and increase the level of apoptosis in cells. A The number of cell apoptosis increased after ELK3 expression was inhibited. B Flow cytometry showed that the apoptosis rate of down-regulated ELK3 cells increased. C Western-blot showed that the expression of apoptotic protein increased after ELK3 inhibition

that inhibiting the expression of ELK3 in gastric cancer Supplementary Information cell lines reduces its proliferation ability and increases its The online version contains supplementary material available at https://​doi.​ apoptosis level. Te above evidence suggested that ELK3 org/​10.​1186/​s12935-​021-​02008-5. may act as an oncogene in gastric cancer, but its specifc mechanism afecting the progression of gastric cancer Additional fle 1: Table S1. Clinical information of the TCGA-STAD cohort. Table S2. Diferentially expressed transcription factors in the TCGA-STAD requires further experimental research. cohort. Table S3. Transcription factors in modules related to gastric can- cer. Table S4. predicts target genes of hub TFs.

Conclusions Acknowledgements In general, we analyzed the expression diferences of tran- We sincerely thank TCGA and GTEx database for the massive shared resources. scription factors in gastric cancer based on public data- Authors’ contributions bases, screened transcription factors with prognostic ability, LZ and ZC: conception, design, data analysis and manuscript writing, YW and used clinical samples for expression verifcation. On and HL: data collection, LX: guide, supervise and review. All authors read and approved the fnal manuscript. this basis, we also constructed a prognostic signature and nomogram and systematically verifed that it has good pre- Funding dictive sensitivity, which is helpful for accurate and person- National Natural Science Foundation of China, Grant/Award Numbers: 81872480, 81760549, 81560492; Natural Science Foundation of Jiangxi alized treatment of gastric cancer. More importantly, we Province, Grant/Award Numbers: 20203BBG73056; Science and Technology have identifed ELK3 as a new biomarker for gastric cancer, Research Project of Education Department of Jiangxi Province, Grant/Award which is benefcial to the precise treatment of gastric cancer. Number: GJJ180024. Zhou et al. Cancer Cell Int (2021) 21:357 Page 17 of 17

Availability of data and materials 17. Guo T, Bai Y, Cheng X, et al. Insulin gene enhancer protein 1 mediates All data can be obtained from the corresponding author’s ofce and public glycolysis and tumorigenesis of gastric cancer through regulating glucose databases. transporter 4. Cancer Commun. 2021;41(3):258–72. 18. Deng R, Zuo C, Li Y, et al. The innate immune efector ISG12a promotes cancer immunity by suppressing the canonical Wnt/β-catenin signaling Declarations pathway. Cell Mol Immunol. 2020;17(11):1163–79. 19. Dongre A, Weinberg RA. New insights into the mechanisms of epithelial– Ethics approval and consent to participate mesenchymal transition and implications for cancer. Nat Rev Mol Cell Biol. All gastric cancer clinical samples involved were approved by the patient’s 2019;20(2):69–84. informed consent and the ethics committee of the Second Afliated Hospital 20. Gonzalez DM, Medici D. Signaling mechanisms of the epithelial–mesenchy- of Nanchang University. mal transition. Sci Signal. 2014;7(344):re8. 21. Im J-Y, Yoon S-H, Kim B-K, et al. DNA damage induced apoptosis suppressor Consent for publication (DDIAS) is upregulated via ERK5/MEF2B signaling and promotes β-catenin- All authors agree to the publication of this article. mediated invasion. Biochim Biophys Acta. 2016;1859(11):1449–58. 22. Scohy S, Gabant P, Van Reeth T, et al. Identifcation of KLF13 and KLF14 Competing interests (SP6), novel members of the SP/XKLF transcription factor family. Genomics. The authors declare that they have no competing interests. 2000;70(1):93–101. 23. Gou D, Wang J, Gao L, et al. Identifcation and functional analysis of a novel Author details 1 human KRAB/C2H2 zinc fnger gene ZNF300. Biochem Biophys Acta. Department of General Surgery, The Second Afliated Hospital of Nanchang 2004;1676(2):203–9. University, 1 Minde Road, Donghu District, Nanchang 330006, Jiangxi, China. 2 24. Xue L, Qiu H, Ma J, et al. ZNF300, a recently identifed human transcription Molecular Medicine Laboratory, The Second Afliated Hospital of Nanchang factor, activates the human IL-2Rβ promoter through the overlapping University, Nanchang 330006, Jiangxi, China. ZNF300/EGR1 binding site. Cell Mol Biol Lett. 2010;15(4):530–40. 25. Wang T, Wang XG, Xu JH, et al. Overexpression of the human ZNF300 gene Received: 7 April 2021 Accepted: 3 June 2021 enhances growth and metastasis of cancer cells through activating NF-kB pathway. J Cell Mol Med. 2012;16(5):1134–45. 26. Yu S, Ao Z, Wu Y, et al. ZNF300 promotes chemoresistance and aggressive behaviour in non-small-cell lung cancer. Cell Prolif. 2020;53(11):e12924. 27. Ahmad A, Zhang W, Wu M, et al. Tumor-suppressive miRNA-135a inhibits References breast cancer cell proliferation by targeting ELK1 and ELK3 oncogenes. 1. Sung H, Ferlay J, Siegel RL, et al. Global cancer statistics 2020: GLOBOCAN Genes Genom. 2018;40(3):243–51. estimates of incidence and mortality worldwide for 36 cancers in 185 coun- 28. Giovane A, Pintzas A, Maira S, et al. Net, a new ets transcription factor that is tries. CA Cancer J Clin. 2021;71(3):209–49. activated by Ras. Genes Dev. 1994;8(13):1502–13. 2. Li Z, Lü M, Zhou Y, et al. Role of long non-coding RNAs in the chem- 29. Ducret C, Maira SM, Lutz Y, et al. The ternary complex factor Net contains oresistance of gastric cancer: a systematic review. OncoTargets Ther. two distinct elements that mediate diferent responses to MAP kinase 2021;14:503–18. signalling cascades. Oncogene. 2000;19(44):5063–72. 3. Latchman DS. Transcription factors: an overview. Int J Biochem Cell Biol. 30. Oh N, Park J, Park J, et al. The role of ELK3 to regulate peritumoral lymphangi- 1997;29(12):1305–12. ogenesis and VEGF-C production in triple negative breast cancer cells. 4. Lee TI, Young RA. Transcription of eukaryotic protein-coding genes. Annu Biochem Biophys Res Commun. 2017;484(4):896–902. Rev Genet. 2000;34:77–137. 31. Lee J, Hur W, Hong S, et al. ELK3 promotes the migration and invasion of 5. Inokuchi S, Aoyama T, Miura K, et al. Disruption of TAK1 in hepatocytes liver cancer stem cells by targeting HIF-1α. Oncol Rep. 2017;37(2):813–22. causes hepatic injury, infammation, fbrosis, and carcinogenesis. Proc Natl 32. Sloan K, Marquez H, Li J, et al. Increased PEA3/E1AF and decreased Net/Elk- Acad Sci USA. 2010;107(2):844–9. 3, both ETS proteins, characterize human NSCLC progression and regulate 6. Bettermann K, Vucur M, Haybaeck J, et al. TAK1 suppresses a NEMO-depend- caveolin-1 transcription in Calu-1 and NCI-H23 NSCLC cell lines. Carcinogen- ent but NF-kappaB-independent pathway to liver cancer. Cancer Cell. esis. 2009;30(8):1433–42. 2010;17(5):481–96. 33. Yoo S, Lee C, An H, et al. RSK2-mediated ELK3 activation enhances cell trans- 7. Lambert SA, Jolma A, Campitelli LF, et al. The human transcription factors. formation and breast cancer cell growth by regulation of c-fos promoter Cell. 2018;172(4):650–65. activity. Int J Mol Sci. 2019;20(8):1994. 8. Ritchie ME, Phipson B, Wu D, et al. limma powers diferential expression 34. Mao Y, Li W, Hua B, et al. Silencing of ELK3 induces S-M phase arrest and analyses for RNA-sequencing and microarray studies. Nucleic Acids Res. apoptosis and upregulates SERPINE1 expression reducing migration in 2015;43(7):e47. prostate cancer cells. Biomed Res Int. 2020;2020:2406159. 9. Langfelder P, Horvath S. WGCNA: an R package for weighted correlation 35. Park J, Kim K, Ko J, et al. PI3K/Akt/mTOR activation by suppression of ELK3 network analysis. BMC Bioinform. 2008;9:559. mediates chemosensitivity of MDA-MB-231 cells to doxorubicin by inhibit- 10. Lu Y, Zhou Y, Qu W, et al. A Lasso regression model for the construction of ing autophagy. Biochem Biophys Res Commun. 2016;477(2):277–82. microRNA-target regulatory networks. Bioinformatics. 2011;27(17):2406–13. 36. Wang S, Li J, Yang X. Long non-coding RNA LINC00525 promotes the 11. Heagerty PJ, Lumley T, Pepe MS. Time-dependent ROC curves for censored stemness and chemoresistance of colorectal cancer by targeting miR-507/ survival data and a diagnostic marker. Biometrics. 2000;56(2):337–44. ELK3 axis. Int J Stem Cells. 2019;12(2):347–59. 12. Alba AC, Agoritsas T, Walsh M, et al. Discrimination and calibration of 37. Li TZ, Kim SM, Hur W, et al. Elk-3 contributes to the progression of liver clinical prediction models: users’ guides to the medical literature. JAMA. fbrosis by regulating the epithelial–mesenchymal transition. Gut Liver. 2017;318(14):1377–84. 2017;11(1):102–11. 13. Kolmykov S, Yevshin I, Kulyashov M, et al. GTRD: an integrated view of 38. Kim KS, Kim J, Oh N, et al. ELK3-GATA3 axis modulates MDA-MB-231 metas- transcription regulation. Nucleic Acids Res. 2021;49:D104–11. tasis by regulating cell–cell adhesion-related genes. Biochem Biophys Res 14. Uhlen M, Zhang C, Lee S, et al. A pathology atlas of the human cancer Commun. 2018;498(3):509–15. transcriptome. Science. 2017. https://​doi.​org/​10.​1126/​scien​ce.​aan25​07. 15. Daly M. Transcription factor defects causing platelet disorders. Blood Rev. 2017;31(1):1–10. Publisher’s Note 16. Wang N, Chen Y, Shi C, et al. CREB3L4 promotes angiogenesis and tumor Springer Nature remains neutral with regard to jurisdictional claims in pub- progression in gastric cancer through regulating VEGFA expression. Cancer lished maps and institutional afliations. Gene Ther. 2021. https://​doi.​org/​10.​1038/​s41417-​021-​00305-9.